Better lambdas!!!!!!!!!!

[Pierre Krafft via Digitalmars-d]

On Saturday, 12 September 2015 at 03:32:51 UTC, Prudence wrote:
> On Saturday, 12 September 2015 at 02:13:11 UTC, Pierre Krafft 
> wrote:
>> On Saturday, 12 September 2015 at 01:03:54 UTC, Prudence wrote:
>>> On Thursday, 10 September 2015 at 18:02:36 UTC, Ali Çehreli 
>>> wrote:
>>>> On 09/10/2015 10:55 AM, Prudence wrote:
>>>> > How bout this:
>>>> >
>>>> > void myfunc(double delegate(int i, int z, float f)) {....}
>>>> >
>>>> >
>>>> > myfunc((int i, int z, float f) { return i*z*f; } }
>>>> >
>>>> > vs
>>>> >
>>>> > myfunc({ return i*z*f; })   // Names of parameters are
>>>> inferred from
>>>> > signature.
>>>>
>>>> Considering other features of the language, that's pretty 
>>>> much impossible in D. What if there is another i in scope:
>>>>
>>>> int i;
>>>> myfunc({ return i*z*f; });
>>>>
>>>> Now, should it call another overload of myfunc that takes 
>>>> (int z, int f) because i is something else?
>>>>
>>>> Should the compiler analyze the body of the code and decide 
>>>> which symbols could be parameters? And then go through all 
>>>> overloads of myfunc? etc.?
>>>>
>>>> Ali
>>>
>>> As I said, it could throw a warning or error. It, in some 
>>> sense, is already a a problem with nested blocks that hide 
>>> outside variables, is it not?
>>>
>>> The compiler doesn't need to scan anything. It knows the 
>>> which parameters from the definition!
>>>
>>>
>>> -> void myfunc(double delegate(int i, int z, float f))  <- 
>>> Compiler knows to use the names here as the default names in 
>>> for the parameters when.
>>>
>>>
>>> when used:
>>>
>>> myfunc({ return i*z*f; }); <- Oh, there are the names, we 
>>> know what they are because the signature is tells us. The 
>>> compiler does the following:
>>>
>>>
>>> 1. Sees we have a block without any parameters defined. i.e., 
>>> a lambda.
>>>
>>> 2. It looks up the signature of myfunc to find out what the 
>>> names are
>>>
>>> 3. It sees that they are i z and f
>>>
>>> 4. Now it knows and it effectively rewrites the code as
>>>
>>> myfunc((i,z,f) { return i*z*f; });
>>>
>>> Surely this is not difficult, 4 steps?
>>
>> You're making your code more brittle for a small gain. The 
>> suggestion makes parameter usage order important and the 
>> compiler can't warn about my typos.
>> Consider:
>> myfunc({return "x:"~x~"y:"-y;}) getting changed to 
>> myfunc({return "y:"~y~"x:"~x;});
>> Or the typo in
>> myfunc({return i*z+f*j;});
>>
>> Lambdas are already very concise. This proposal doesn't give 
>> any benefits outside of very simple lambdas. Such lambdas are 
>> already so simple that they could use some standard functions 
>> instead (like sum, to!T, and bind).
>
>
> What does this have to do with my proposal? Those issues exist 
> regardless of the simplification.
>
> myfunc({return "x:"~x~"y:"-y;}) getting changed to
> myfunc({return "y:"~y~"x:"~x;});
>
> huh? What do you mean the suggestion makes parameter usage 
> order important? They are important, it has nothing to do with 
> the suggestion? Are you saying that you want to reserve the 
> right to do something like
>
> myfunc(string delegate(string x, string y));
>
> and
>
> myfunc((y,x){ "y:"~y~"x:"~x; })
>
> ? If so, or unless I'm missing something, that's bad no matter 
> what. Changing the order and reversing the names is more than 
> just confusing, it's hard to read and most people will gloss 
> over that fact. Be consistent with your parameters and maybe 
> you'll have less bugs?
>
>
>
> Or the typo in
>
> myfunc({return i*z+f*j;});
>
> Again, what does this have to do with anything? A typo is a 
> typo and is always a mistake. The above example has the same 
> effect regardless if the parameters are explicit or deduced.
>
>
> myfunc((i,z,f) {return i*z+f*j;});
>
> j is still a problem. If j is defined outside the lambda then 
> regardless of specific or implicit parameter names, it will not 
> cause any problems.
>
> In either case, the compiler can see that j is either 
> referenced outside the scope or undefined. It has nothing to do 
> with the parameters used.
>
>
> Of course maybe I'm missing something, but essentially are not 
> almost all uses of lambda's simply copying the parameter 
> signature of the delegate. It already infers types... you could 
> say that leads to typo's too...

  myfunc({return "x:"~x~"y:"-y;});
is infered to mean myfunc((x,y){return "x:"~x~"y:"-y;});
while
  myfunc({return "y:"~y~"x:"~x;});
is infered to mean myfunc((y,x){return "y:"~y~"x:"~x;});
which is not what I expect since the lambda I want is 
myfunc((x,y){return "y:"~y~"x:"~x;});
This can lead to subtle bugs which are very hard to see.

In the typo example there could be two overloaded functions 
differing only in that one takes a delegate having 3 parameters 
and one taking a delegate having 4 parameters. If we explicitly 
write the lambda parameters the typo will be found since j is 
undefined. But when parameters are inferred the compiler will see 
that {return i*z + f*j;} matches the function taking a lambda 
with 4 parameters.

Inferred parameter types are on the brink of what I can allow. 
They can risk typos, but not as easily since you write the 
parameters twice (declaration and usage). They can also silently 
change if the function taking the delegate has the parameter type 
changed. I don't want to add more magic to that area.